An incremental learning of neural network with multiplication units for function approximation

Dazi Li, Kotaro Hirasawa, Takayuki Furuzuki, Kiyoshi Wada

Research output: Contribution to journalArticle

Abstract

This paper presents a constructive neural network with sigmoidal units and multiplication units, which can uniformly approximate any continuous function on a compact set in multi-dimensional input space. This network provides a more efficient and regular architecture compared to existing higher-order feedforward networks while maintaining their fast learning property. Proposed network provides a natural mechanism for incremental network growth. Simulation results on function approximation problem are given to highlight the capability of the proposed network. In particular, self-organizing process with RasID learning algorithm developed for the network is shown to yield smooth generation and steady learning.

Original languageEnglish
Pages (from-to)135-140
Number of pages6
JournalResearch Reports on Information Science and Electrical Engineering of Kyushu University
Volume8
Issue number2
Publication statusPublished - 2003 Sep

Fingerprint

Neural networks
Learning algorithms

Keywords

  • Function approximation
  • Higher order neural networks
  • Multiplication units
  • Random search
  • Sigmoidal units

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Hardware and Architecture
  • Engineering (miscellaneous)

Cite this

An incremental learning of neural network with multiplication units for function approximation. / Li, Dazi; Hirasawa, Kotaro; Furuzuki, Takayuki; Wada, Kiyoshi.

In: Research Reports on Information Science and Electrical Engineering of Kyushu University, Vol. 8, No. 2, 09.2003, p. 135-140.

Research output: Contribution to journalArticle

@article{35390aa0fc4a41b380ece14ea2abc602,
title = "An incremental learning of neural network with multiplication units for function approximation",
abstract = "This paper presents a constructive neural network with sigmoidal units and multiplication units, which can uniformly approximate any continuous function on a compact set in multi-dimensional input space. This network provides a more efficient and regular architecture compared to existing higher-order feedforward networks while maintaining their fast learning property. Proposed network provides a natural mechanism for incremental network growth. Simulation results on function approximation problem are given to highlight the capability of the proposed network. In particular, self-organizing process with RasID learning algorithm developed for the network is shown to yield smooth generation and steady learning.",
keywords = "Function approximation, Higher order neural networks, Multiplication units, Random search, Sigmoidal units",
author = "Dazi Li and Kotaro Hirasawa and Takayuki Furuzuki and Kiyoshi Wada",
year = "2003",
month = "9",
language = "English",
volume = "8",
pages = "135--140",
journal = "Research Reports on Information Science and Electrical Engineering of Kyushu University",
issn = "1342-3819",
publisher = "Kyushu University, Faculty of Science",
number = "2",

}

TY - JOUR

T1 - An incremental learning of neural network with multiplication units for function approximation

AU - Li, Dazi

AU - Hirasawa, Kotaro

AU - Furuzuki, Takayuki

AU - Wada, Kiyoshi

PY - 2003/9

Y1 - 2003/9

N2 - This paper presents a constructive neural network with sigmoidal units and multiplication units, which can uniformly approximate any continuous function on a compact set in multi-dimensional input space. This network provides a more efficient and regular architecture compared to existing higher-order feedforward networks while maintaining their fast learning property. Proposed network provides a natural mechanism for incremental network growth. Simulation results on function approximation problem are given to highlight the capability of the proposed network. In particular, self-organizing process with RasID learning algorithm developed for the network is shown to yield smooth generation and steady learning.

AB - This paper presents a constructive neural network with sigmoidal units and multiplication units, which can uniformly approximate any continuous function on a compact set in multi-dimensional input space. This network provides a more efficient and regular architecture compared to existing higher-order feedforward networks while maintaining their fast learning property. Proposed network provides a natural mechanism for incremental network growth. Simulation results on function approximation problem are given to highlight the capability of the proposed network. In particular, self-organizing process with RasID learning algorithm developed for the network is shown to yield smooth generation and steady learning.

KW - Function approximation

KW - Higher order neural networks

KW - Multiplication units

KW - Random search

KW - Sigmoidal units

UR - http://www.scopus.com/inward/record.url?scp=0347024904&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0347024904&partnerID=8YFLogxK

M3 - Article

VL - 8

SP - 135

EP - 140

JO - Research Reports on Information Science and Electrical Engineering of Kyushu University

JF - Research Reports on Information Science and Electrical Engineering of Kyushu University

SN - 1342-3819

IS - 2

ER -